基于高斯加权和流形的高保真彩色图像降噪

Color image denoising based on Gaussian weighting and manifold for high fidelity

针对用矢量法对彩色图像进行降噪处理,算法复杂度较高,无法达到实时处理的问题,提出了基于改进高斯加权和自适应流形的高保真彩色图像降噪方法。首先,将彩色图像用非局部均值算法得到高维数据,使用改进的高斯内核对彩色图像进行加权计算;然后,采用抛雪球方法处理这些高维数据,以高斯距离为权值,投影每个像素点的颜色到自适应流形;接着,对流形进行平滑降维,采用迭代法实现图像平滑;最后,收集流形中的平滑值,将平滑值对所有像素进行插值,得到降噪后的图像数据。实验证明,该方法对彩色图像进行降噪处理后,能够很好地保留原图像的细节,不会掺杂周围像素的颜色,算法处理速度较快,能够达到实时处理效果,降噪效果与原算法相比峰值信噪比(PSNR)提高近2.0 dB,结构相似度提高了1百分点以上。

Using vector method for color image denoising, the complexity of the algorithm is high and cannot achieve real- time performance. A method for high fidelity color image denoising was proposed based on Gaussian weighting and adaptive manifold. Firstly, it used the non-local means algorithm to get high-dimensional data, and used the improved Gaussian kernel to calculate the weight of color image. Secondly, splatting method was used to deal with the high-dimensional data, and a Gaussian distance-weighted projection of the colors of all pixels was performed onto each adaptive manifold. Thirdly, smooth dimensionality reduction was done on convection shape, and iterative method was used for image smoothing. Finally, the final filter response was computed for each pixel by interpolating blurred values gathered from all adaptive manifolds. The experimental results show that the algorithm has a superior denoising performance than the original one, and it also can improve real-time performance. By using this algorithm, the details can be preserved well. Peak Signal-to-Noise Ratio （PSNR） can be improved nearly 2.0 dB, and Structural Similarity Index Measurement （SSIM） can be improved more than 1%.